Landscape Area Measurements for Urban Outdoor Water Budgets

Introduction

In 2018, the state of California passed legislation related to water conservation and drought planning. This legislation will require all urban water retailers to meet state-defined annual water use targets. To meet these objectives, retailers will need to more closely track the different components of their water use budgets (e.g. residential indoor, residential outdoor, commercial outdoor, etc.).

This project will focus on commercial outdoor water budgets. For the purposes of meeting state water use objectives, a retailer only needs to consider commercial properties (e.g. parks, golf courses, universities, hospitals) which have a water meter dedicated entirely to irrigation. If a commercial property has a large landscaped area, it is more likely to have a dedicated irrigation meter. All other commercial outdoor water use can be ignored.

The outdoor budget for these commercial properties is calculated by multiplying the total outdoor landscaped area of these properties by a standard evapotranspiration (ET) value. This ET value is set by the state based on their definition of "efficient water use." It is up to the retailer to determine their total commercial outdoor landscaped area to insert into the equation.

This project will outline an approach for retailers to remotely calculate the total landscaped area of commercial properties using classified 1 meter resolution NAIP imagery, county parcel data, and point locations of irrigation meters.

The major challenge in this process will be the land cover classification of the NAIP imagery to identify vegetation pixels. Details on the land cover classification approach can be found in section 1.5.

Study area

The approach will initially be developed for a small area within the greater Long Beach service area. Once the individual steps have been successfully established, the workflow can then be scaled to analyze an entire service area. The location of the Long Beach service area is shown below.

NAIP imagery

This approach will use 2020 NAIP data. For the initial workflow development, only 1 NAIP tile will be used. Each individual image covers a 3.75 x 3.75 minute quarter quadrangle plus a 300 meter buffer on all four sides. This is a small subset of the Long Beach service area, as shown below.

Notice that the tile doesn't entirely overlap the Long Beach service area. For the purposes of this exercise, we will do the analysis on the entire NAIP test tile so that we have more area to work with. In the future, we can crop the final shapefile of the classified properties to the service area boundary.

Notice above that there are several properties with large landscaped areas. Golf courses, a park, and sports fields are identifiable from this view. These are the landscaped areas we are interested in identifying and calculating their associated areas.

Overlaying commercial parcel boundaries

LA County maintains a shapefile of all parcels in the county. This file also contains information about the property, including whether it is a commercial or residential property. We can overlay a shapefile of commercial property boundaries onto our test tile image.

I finally got the crazy LA County ArcGIS portal to spit out some parcel data, but it seems to have only exported those within the extent of the Online Map Viewer. I then had to open the layer in ArcGIS Pro desktop, then export it as a shapefile to import here. I'll work on this more.

This cii_parcel shapefile will eventually be used to calculate the landscaped area once the NAIP imagery is classified.

Land cover classification overview

For the land cover classification, I will use the RandomForest (Brieman 2001) ensemble decision tree algorithm by Leo Breiman and Adele Cutler, outlined in the random-forest-classification tutorial. RandomForest utilizes the scikit-learn Python library.

This is a supervised classification approach which inputs training polygons. The training polygons were manually created in ArcGIS Pro using the NAIP test tile as guidance, and assigned a classification of either 1-Turf or 2-Shrubs/Trees.

Next steps

The training polygons can now be incorporated into the RandomForest workflow. Next steps include: